Continu ous Time Goes by* Russell

Continu ous Time Goes by* Russell Uwe Lück January 25, 2004 Abstract Russell and Walker proposed different ways of constructing instants from events. As an explanation of time as a continuum, Thomason favoured Walker s construction. The present paper shows that Russell s construction fares as well. To this end, a mathematical characterization problem is solved which corresponds to the characterization problem that Thomason solved with regard to Walker s construction. It is shown how to characterize those event structures (formally: interval orders) which, through Russell s construction of instants, become linear orders isomorphic to a given (or, deriving: to some non-trivial ordered) real interval. As tools, separate characterizations for each of resulting (i) Dedekind completeness, (ii) separability, (iii) plurality of elements, (iv) existence of certain endpoints are provided. Denseness is characterized to replace Russell s erroneous attempt. Somewhat minimal non-constructive principles needed are exhibited. Keywords: time, Russell, instants from events, continuum, interval orders, axiom of choice. *The song by H. Hupfeld, played in the movie Casablanca. Seminar für Philosophie, Logik und Wissenschaftstheorie; Philosophie-Department; Universität München; Geschwister-Scholl-Platz 1; 80539 München; Germany; uwe.lueck @lrz.uni-muenchen.de. This paper idealizes the content of my talk entitled Exactly which orders of periods crystallize to continua of time points? at the conference of October 25, 1999, at the University of Konstanz organized by Volker Halbach. The subject was brought to my attention by my teachers Professor Godehard Link and Karl-Georg Niebergall, and Volker Halbach s invitation initiated my search for the solutions presented here. The paper underwent changes following suggestions by Volker Halbach, Peter Schuster, and Nick Silich (the latter being expert for questions of English). Louis Narens coauthored the abstract. Roland Kastler, Hannes Leitgeb, and Marek Polanski encouraged me in pursuing the matter and writing the paper. The paper has been typeset by TEX using L A TEX 2ε and the proof environment from AMS-L A TEX; Alexander Gröpl made the only text editor I like to use for typing TEX-files. Conceiving and typing the paper would not have been possible without substantial financial support by my mother, Mrs. Renate Lück. I am very grateful to all of them for what they have done and been as I told above, as well to the other friendly attendands of the mini-conference. A revision of the paper was enabled by an exhibition according to the Bavarian Law for the Advancement of Young Scientists and Artists dating from December 18, 1984. 1

2 1 INTRODUCTION AND SUMMARY. Contents 1 Introduction and summary. 2 2 Relations, linearity, and real intervals. 4 3 Some events, some Russellian instants. 5 4 Core Task. 6 5 Via completeness and dense subsets. 8 6 Pivotal derived notions. 10 7 Solutions. 13 8 Proofs. 18 8.1 Constructivity as opposed to choice principles........ 18 8.2 Existence of Russellian instants of time............. 20 8.3 Derived relations, witnesses, and linearity of time....... 24 8.4 Further notation and Theorem 1................. 26 8.5 First and last instants; Theorem 2................ 28 8.6 Dense subsets; Theorem 3, Corollary 1............. 30 8.7 Completeness, Theorem 4..................... 34 9 Philosophical and historical remarks. 37 9.1 Russell on continuity of time and on conscious events required. 37 9.2 Historical understandings of continuum............ 39 9.3 Continuity vs. separability and completeness.......... 41 9.4 Relative merits of Russell s vs. Walker s construction..... 42 9.5 In metaphysicam........................ 43 9.6 Event relations beyond interval orders.............. 44 1 Introduction and summary. The problem the present paper is concerned with derives from Russell s [42, Lecture IV] attempt mathematically to construct instants of time from events. Thomason s [46] paper starts almost the same way. According to Thomason, however, Russell s construction has the disadvantage that it is difficult to see what assumptions about the temporal relationships among events will ensure that the instants constructed comprise a continuum, isomorphic to the real numbers. Thomason [46] thus presents an alternative way due to the physicist A. G. Walker 1 of constructing instants of time out of events. Thomason shows of some conditions on the temporal relationships of events that they do ensure instants of time, as constructed 1 For hints on Walker s inheritance in physics, see footnote 16.

3 according to Walker, comprising a continuum. 2 He then concludes that Walker s theory offers, as Russell s does not, a plausible explanation of time as a continuum. The present paper shows (besides some by-products) how to single out those relations on sets of events which become order-isomorphic to the real numbers through Russell s construction. More generally, it shows how to tell from such a relation whether the resulting order is in a class (closed under isomorphisms) of intervals of the real numbers 3 an interval being open in each, one, or no direction; ordered by the relation of being less. 4 At least one of the characterizations presented might be considered a refutation of Thomason s claim denying explanatory power of Russell s theory. 5 The keys to these characterizations are (i) a sufficient condition on events and their temporal relationships for Russellian time being dense 6, due to N. Wiener [49, pp. 446f.] or to Russell; 7 (ii) a characterization, due to Russell, of those events that get first or, resp., last instants by his construction. The sufficient condition according to (i) is modified here to obtain a necessary and suffient one. The main results are stated in Section 7. The sections which precede Section 7 merely explain the notions used in the latter. In proving claims afterwards, I keep books on what non-constructive choice or maximality principles (the very axiom of choice or something weaker) I use. It will even be shown that one of the characterizing conditions can be used as an alternative to any such principle in the relevant context. Some proofs merely replace existing Principia-notation proofs and thus may be helpful to at least some readers. The last section comments on some remaining aspects of continuumlikeness of time. 2 Indeed, the conditions he presents are necessary as well. Kleinknecht [26] does something very similar, but presents a pair of conditions which is only sufficient, not necessary. 3 A real interval as defined below, i.e. 4 This extends Thomason s and Walker s scope to something of which admittedly physicists will hardly acknowledge any use of see Section 4 below. 5 Subsection 9.4 will discuss this statement. Maybe even Thomason s difficulty claim is refuted. 6 As defined in Corollary 1 below. 7 I have had some troubles in distinguishing Russell s from N. Wiener s credits. According to the footnote of [49, p. 441], Wiener (the well-known mathematician who later founded cybernetics ) investigated the matter on Russell s suggestion. Indeed, at that time Wiener was a student under Russell at Cambridge University ([31, pp. 45ff.], [18]). Thus, while Wiener explicitly attributes the definition of instants under consideration here and another notion to Russell, it is no surprise when further credits are difficult to track. Moreover, the first edition of Russell s [42] appeared in the same year as Wiener s [49].

4 2 RELATIONS, LINEARITY, AND REAL INTERVALS. 2 Relations, linearity, and real intervals. Already in explaining my goal I will come across several binary relations; so I declare my general conventions concerning them in advance. When X, Y are any sets and R X Y is some binary relation between them, I will write x R y instead of x, y R. Considering some number R 1,..., R n of binary relations, I will write x 1 R 1 x 2 R 2 x 3... x n R n x n+1 meaning that x 1 R 1 x 2, x 2 R 2 x 3,... and x n R n x n+1. 8 A related set is an ordered pair X, R where R is a binary relation on X. 9 If R is no subset of X X, X, R is shorthand for X, R (X X). Sometimes I will abuse language by talking of elements and cardinality of related sets X, R as if I would talk of X. Related sets X, R, X, R are isomorphic (to each other), if there is a one-to-one map φ from X onto X such that for any x, y X x R y if and only if φ(x) R φ(y). A least element in a related set X, R is an x X such that x R y for any other y X. If instead y R x, then x is a greatest element. By a (strict) linear order on some set Y I understand a transitive binary relation R on Y such that for any x, y Y exactly one of x R y, y R x and x = y holds (this mixes irreflexivity, connectedness, and even, redundantly, asymmetry). 10 In this case, I will call any related set X, R such that X Y a linearly ordered set, or, for short, a loset. 11 R, < will denote the set of all real numbers linearly ordered by the binary relation of being less. By a real interval I will understand a set I R such that if s, t I and s < r < t then r I. I am going to call a real interval open if it contains neither a least nor a greatest element, half-open if it contains a least or a greatest element but not 8 Formulas like these and others are just meant to be shorthand for English mathematical expressions; they are not meant to refer to expressions of a formal language from the realm of metamathematics. However, metamathematical questions may rise eventually, and one will then be able easily to realize that some condition discussed can be formalized in some first-order language, e.g. 9 R is a binary relation on X, if it is one between X and X. 10 Now that [40] is a relevant reference, it should be noted that transitivity here as will be usual for many readers means that x R z whenever x R y and y R z; it should not be confused with what in [40] and other contexts is denoted by the same term. In these contexts, transitive could have been replaced by homogeneous. By the trichotomy condition, the union of domain and codomain of R is the only set on which a given binary relation R can be a linear order. 11 While poset is a standard term in the literature on partially ordered sets, I have never seen this term loset. For didactical reasons, however, I urgently need a short term. Chain is short as well, but should work like antichain and therefore does not fit.

5 both, and compact if it contains both a least and a greatest element (which may coincide). By a non-trivial real interval I understand a real interval having more than one element which is the same as having the cardinality of R. Observe that being non-trivial is the same as being non-void in case of an open real interval and as having distinct borders in the case of a compact real interval, while all half-open intervals are non-trivial. 3 Some events, some Russellian instants. For the remaining, I fix some non-void set E of so-called events and a binary relation P E E on it, the philosophical meaning of which is to be wholly precedes. Proofs of what is claimed on the few lines following are indicated in the first two subsections of Section 8. An antichain (in E, P ) is a subset A of E such that for any two a, b A neither a P b nor b P a. A maximal antichain is an antichain that cannot be extended to another antichain by adding any non-void set of events. By the axiom of choice, any antichain can be extended to a maximal antichain, and as the empty set is an antichain, there is at least some maximal antichain. For the same time that E, P is fixed, let L be just the set of maximal antichains in E, P. By Russell s proposal, 12 the elements of L are instants of time as constructed from those events presented in E and from their temporal relationships encoded by P. (Therefore, it will be informal variables like s, t that range in L, while A, B,... will range over arbitrary subsets of E.) The order of time, then, is defined to be T := { s, t L L a P b for some a s and some b t }. In order notation and philosophical diction, s T t (only) for instants s, t in L just if some event a in s wholly precedes some event b in t. I furthermore fix that P be irreflexive (i.e., no event precede itself) and that for any events a, b, c, d, if a P b and c P d then a P d or c P b. (I) ( E, P is now an interval order in the sense of [16] and [17]; and I may use results from Wiener [49], 13 who calls a P satisfying an equivalent pair of conditions a relation of complete sequence, as well as from Russell s [43], who uses a triple of conditions equivalent to each of the previously mentioned pairs.) Reading (I) as if a wholly precedes b and c wholly precedes d, but c does not wholly precede b, then a wholly precedes d may exhibit (I) to 12 For distinguishing Russell s from A. N. Whitehead s credits concerning constructing points and instants, cf. [1, pp. 252f.]. 13 Wiener s paper is summarized in modern notation by [18].

6 4 CORE TASK. be a self-evident feature of P if the latter is read as wholly precedes. 14 This would render E, P quite a good starting point for philosophically justifying instants of time. In fact, in the situation of the hypothesis of the proposed reading, (i) either b wholly precedes c, and then by selfevident transitivity of wholly precedes a wholly precedes d; (ii) or b and c have something in common which must be preceded by a and precede d, and this might be considered some evidence for a wholly preceding d. 15 By (I) and irreflexivity of P, the latter relation is transitive. Moreover, (I) ensures that T is a linear order on L [49, p. 445f.]. In this respect, T is an adequate mathematical reconstruction of temporal precedence between instants of time. Some aspects of adequacy, however, remain to be considered. 4 Core Task. I am going to propose a few characterization problems to be solved, indicating them as questions. First compare the characterization problem that Thomason [46] deals with. He explains how Walker would construct, instead of L, T, another time ordering L, T from E, P. He then presents a necessary and sufficient condition on E, P for L, T being isomorphic to R, <. The anologous problem for L, T he considers difficult, viz., Question 1. When, in terms of E, P, is L, T isomorphic to R, <? Here, I am using when... to abridge something like What conditions on E, P are necessary and sufficient for E, P being isomorphic to R, <?, and similarly below. These questions are, admittedly, not perfectly precise, but I am not attempting to explicate the notion of characterization in this paper; it will suffice to present convincing solutions to the problems, however ill-posed the latter may be. We will encounter solutions for other characterization problems by conditions which could be formalized as firstorder sentences in a language interpretable in E, P. It may be no surprise in connection with the continuum, however, that the main characterizing conditions are not first-order, but stipulate a countable set of objects from E, P having a certain property (cf., e.g., [35, Theorem 2.1]). Variants of Question 1 may be interesting from a physical, mathematical, or perhaps even philosophical point of view. First, one might really want an explanation of time as a contiuum, but one might disagree with 14 This is, in only a slightly different situation, held by [1, p. 252]; also cf. [30, p. 181]. 15 Cf. [30, p. 181]. For further interpretations or applications of interval orders, cf. [17, pp. 20ff.]. In philosophically justifying instants of time, however, it is difficult to say what it is that b and c have in common in the second case without committing a petitio principii. I am afraid the only convincing motivations for (I) refer to how interval structures are represented by point structures like the present L, T.

7 Thomason as to what a continuum is. Indeed, the purportedly well-known order-type of the continuum that Cantor characterized in his celebrated [11] was actually that of the closed (i.e., compact) unit interval, and another notion of continua, entailing topological compactness as well, was prominent in a branch of early topology, so-called curve theory (cf. Subsection 9.2). So (as some readers conclude immediately and others will see soon) a continuum may be and has been considered something essentially different to a structure comprising all the real numbers. Unlike the real numbers, it might have a first point, a last point, or both. On the other hand, the question of what a or the continuum is might be considered too theoretical, namely only of mathematical or even philosophical interest. It might be considered more important whether the construction renders time as it is used in physics. Indeed, physical textbooks seem to postulate that instants of time are just the same as real numbers. However, from the viewpoint of physical, relativistic cosmology (of General Relativity, i.e.) which prevails nowadays, astronomical observations indicate that not every real number corresponds to an instant of time (in the usual way): a Big Bang is believed to have started time; and a final collapse of the universe is reckoned with to end time. 16 However, this reasoning may only lead to consider bounded open real intervals instead of all the real numbers, and this switch does not change anything at least order-theoretically or topologically. First or last instants which would make a difference seem to play no role in physics you usually encounter open time intervals in textbooks, even the Big Bang has no starting instant of time. 17 Yet, some physicist might one day leave the herd. So there may be some, if only little, reason to consider more general versions of Question 1. The additional answers will not need extra mathematical effort. From the results the reader may choose what he likes most. The 16 This, however, stems from the special role time is given in the mathematical rendering of how the universe seems to behave, and talking of time perhaps only makes sense when spacetime is a product of time and space in some special sense (I do not expect spacetime singularities of black holes that are formed eventually or that dissolve by a reasoning of Stephen Hawking [21] would deprive the picture of one time line for all point events of spacetime of use). Of course, in some metaphysical view Time could have existed earlier than the Big Bang and could since then somehow interplay with mathematicalphysical time. Such a metaphysical everexisting Time would be of no empirical value, only, maybe, of some aesthetical one if someone s taste behaves appropriately. What is most important here: such a metaphysical Time outside of physical time could not be reconstructed from empirical events! Empirical events occur after the Big Bang and before a final collapse and will therefore only detect the instants of bounded physical time. Indeed, Walker s approach presented by [46] was part of justifying, by assumptions about simple observations, a mathematical model of the whole development of the universe according to General Relativity (cf. references in [46]). The spacetimes according to this mathematical model are nowadays called Robertson Walker spacetimes (cf. [37, pp. 341ff.], e.g.). 17 Cf. footnote 16.

8 5 VIA COMPLETENESS AND DENSE SUBSETS. reader may like to know something about continua, and she may conceive of continua so that two of them are isomorphic to each other (like the open real intervals), or so that there are two or more isomorphism types of continua. E.g., a continuum might be viewed to be anything order-isomorphic to some non-trivial real interval. Consider Question-Scheme 1*. When, in terms of E, P, is L, T isomorphic to I, <? This scheme produces one question for each real interval I. Question 1 is that example where I = R. The scheme represents readers with somewhat very narrow conceptions of continuumlikeness. As for an example of a broad conception, its answers enable answering Question 1 +. When, in terms of E, P, is L, T isomorphic to some I, < where I is a non-trivial real interval? 5 Via completeness and dense subsets. The following is well known: 18 Fact 1. A related set is isomorphic to R, < if and only if it is a non-void, complete and separable loset having neither a least nor a greatest element. To understand this, recall some definitions. Let X, R be some related set again. A lower bound of some Y X is an x X which is a least element of {x} Y (x may be element of Y ). A greatest lower bound of Y X is a greatest element of the set of all lower bounds of Y. By analogy, an upper bound of Y is greatest element of the union of its singleton with Y, and a least upper bound is a least element of the set of all upper bounds. (If X, R is a loset, Y has at most one greatest lower or, resp., least upper bound, of course.) Now, X, R is complete if every non-void subset of X having a lower bound has a greatest such and every subset of X having an upper bound has a least such. 19 A set Y is dense in X, R if Y X and for any two x, z X such that x R z there is some y Y such that x R y R z. 18 According to [46, p. 94] (where denseness is redundant, viz., following from existence of a dense subset); cf. [40, pp. 33 37], where instead of Exercise 2.29 it suffices to think of the set of the rational numbers as a countable set dense in R, <. 19 For the sake of consistency, please forget about what a graph theorist understands by a complete graph. Consistency is only endangered at first sight as long as like in [17, p. 1] complete is used meaning connected when talking of binary relations. But, again, beware of calling a completely ordered set, e.g., complete.

9 Finally, X, R is separable if there is a countable 20 subset of X being dense in X, R. 21 In view of Fact 1, and since L, T is a non-void loset anyway (by Section 3), we approach our goal by moving from Question 1 to Question 1. When, in terms of E, P, is L, T complete and separable without least or greatest element? Question 1 can be splitted into the following questions, which will be dealt with separately from each other. Question 2. When, in terms of E, P, is L, T complete? Question 3. When, in terms of E, P, is L, T separable? Question 4a. When, in terms of E, P, has L, T a least element? Question 4b. When, in terms of E, P, has L, T a greatest element? To deal with Question Scheme 1*, Fact 1 can be generalized to arbitrary real intervals taking some subtleties (as observed in Section 2) into account: Fact 1*. Let I be a real interval. (a) Assume I is open. If I is empty, then I, < =,, and this is the only related set isomorphic to I, <. Otherwise a related set is isomorphic to I, < if and only if it is a non-void complete and separable loset having neither a least nor a greatest element. (b) If I contains a least element, but no greatest, [the other way round, resp.], a related set is isomorphic to I, < if and only if it is a complete and separable loset with 22 a least [greatest, resp.] and without greatest [least, resp.] element. (c) Assume I is compact. 23 If I contains just one element, then a related set is isomorphic to I, < if and only if it is {x}, for some object x. Otherwise a related set is isomorphic to I, < if and only if it is a complete and separable loset with a least and a greatest element where the latter do not coincide. 24 20 By my understanding here, countable does not imply infinite. So some questions of cardinality can be discussed separately below. 21 Cf. [40, Definition 2.28]. 22 So non-void is not needed. 23 My definition slightly differed from the usual topological one, and I is implied here. 24 According to [40, p. 40], the non-trivial cases may be proved essentially the same way as Fact 1. Alternatively, one could derive them from Fact 1 by observing what is preserved when one removes or adds least or greatest elements. The enumeration of order types in [40, p. 40] is imprecise, if not incomplete.

10 6 PIVOTAL DERIVED NOTIONS. So there are four types (with respect to order-isomorphisms) of nontrivial real intervals: open, half-open (two types), and compact-with-morethan-one-element. Questions 2, 3, 4a, and 4b suffice to recognize the first three cases for L, T (corresponding to cases (a) and (b) of Fact 1*) from looking at E, P. For recognizing the fourth case, it suffices to deal with the following final question. Question 5. When, in terms of E, P, has L more than one element? An obvious answer-scheme to Question-Scheme 1* arises, following the lines of Fact 1*. An answer to Question 1 + derives which may be considered a consequence of the following fact, which is entailed by Fact 1*. Fact 1 +. A related set is isomorphic to I, < for some non-trivial real interval I if and only if it is a complete and separable loset having more than one element. 6 Pivotal derived notions. I am going to introduce further relations on the set E of events in order to state some conditions more succinctly than I could do without them. At the same time, some visualizing possible situations may be in order so that it is easier to understand what I mean. Assume for a moment E is a three-element set {a, b, c} and P = { a, b }. This situation is visualized by Figure 1. 25 a P b is visualized by arranging a b c direction of time Figure 1: Three events. horizontal strokes representing a and b so that a vertical stroke can be filled in right-hand to the horizontal stroke representing a and left-hand to the horizontal stroke which represents b. By contrast, no vertical stroke would have two horizontal strokes on different sides such that one of them would represent a and the other would represent c. This holds for b in place of a, as well. Rather, c is overlapping a as well as b. (By the way, {a, c} and {b, c} are antichains; maximal antichains, in fact, and the only ones. So they form the set L of instants of time, and we have {a, c} T {b, c}.) 25 Cf. [46, p. 88].

11 Furthermore, c begins earlier than b, witnessed by a overlapping c but wholly preceding b. Similarly, a ends earlier than c, witnessed by b overlapping c but being wholly preceded by a. The above moment (specializing E, P ) is over, and I generalize the situation by further definitions of binary relations on E representing the relationships observed above. S := { a, b E E neither a P b nor b P a }; SP := { c, b E E c S a P b for some a }; 26 P S := { a, c E E a P b S c for some b }. 27 (Visualize by help of Figure 1.) Now a S b is to mean that a overlaps b, the other way round, or just that a and b overlap, however you like; 28 c SP b is to mean that c begins earlier than b; and a P S c is to mean that a ends earlier than c. 29 An event a will be called an SP -minimal element of a subset A of E, if a A and there is no b A such that b SP a. Dually, 30 an event a will be called a P S-maximal element of A, if a A and there is no b A such that a P S b. 31 The relations on E defined just before make it easier to define the notion of having a first [last, resp.] instant in one line, which in [43] plays in important role for the question of the existence of instants. This notion is vital for formulating answers to all my questions but one as well. To enable the reader to make sense of my definitions following, I precede an outline of Russell s [43] discovery in terms of his philosophical interpretation. Call an event b a contemporary of some event a whenever they overlap (purely mathematically: a S b). Call b an initial contempory of a if additionally a does not begin earlier than b (not a SP b). Now Russell found out 26 If such an a exists, it is an element of E by the definition of a related set or, sufficing as well, by the definition of S. 27 Of course, SP and P S are just the compositions of the binary relations P and S and vice versa, resp., in the sense of [17, p. 3]. 28 Note: a S b if and only if {a, b} is an antichain; and a subset A of E is an antichain if and only if a S b for all a, b A. 29 I am nothing but reporting some notation and terminology introduced by Russell in [43]. E, P, SP, P S is now an event ordering in the sense of Thomason [46, Definition 1]. 30 The notion of duality can by made rather precise, cf. [2, p. 13]. I could have made it quite precise if I had introduced a formal language. Without, just think of the dual of a statement (or a condition ) as the result of interchanging the event symbols on both sides of P, if P is the only relation symbol (besides =) occuring and if conjunctions are written out instead of using chain notation. Using chain notation for conjunctions, consider a one-term conjunction a conjunction chain as well; then the dual is the result of reversing all conjunction chains still if no symbols for derived notions occur. Otherwise, S remains unchanged, while SP is replaced by P S and vice versa, and SP -minimal is replaced by P S-maximal and vice versa. Finally the symbol introduced soon has to be replaced by + and vice versa. 31 SP -minimal and P S-maximal elements may exist because, by Lemma 1 below, SP and (dually) P S are irreflexive.

12 6 PIVOTAL DERIVED NOTIONS. that an event e has a first instant (i.e., there is a T -least maximal antichain containing e) if and only if (keep the following in mind for a moment), whenever e begins earlier than some event a, this a is wholly preceded by some initial contemporary of e. (In this case, the set of initial contemporaries of e is that first instant.) To make mathematical use of the notion of having a first instant, I introduce a symbol denoting its extension (as far as E is concerned): := { e E whenever e SP a, e (S \ SP ) b P a for some b }. So e philosophically means that e has a first instant. 32 Dually, + := { e E whenever a P S e, a P b (S \ P S) e for some b } contains all events having a last instant. (However, I am not quite sure about whether it was really Russell who found that e if and only if e has a first instant, and about what actually Wiener contributed to the result.) 33 32 I prefer writing e (and introducing that notation) to just writing e has a first instant in order to call to the reader s mind that the notion I am making use of is a purely mathematical one. Moreover, to make clear that I am answering questions in terms of E, P in the sense of Section 4), I want to emphasize that I am merely talking of elements e of E such that, e (S \ SP ) b P a for some b whenever e SP a; I am not begging the question by talking about maximal antichains containing e. 33 Russell once claimed the result was his own, but his references indicate the possibility of him only having conjectured it, while it was Wiener who found the proof. To be precise: (i) At the beginning of [43], Russell credits Wiener [49] with having shown what conditions are necessary in order that maximal antichains should form a linearly ordered set as outlined above. By his next sentence, Russell claims he (himself) had shown that every event has a first instant if it satisfies the condition stated above and defining referring to his [42]. A few pages behind, he proves the converse. (ii) Russell s [42, Lecture IV], however, where the subject is discussed, offers no mathematical proof of anything. Instead, one paragraph only claims [i]t will be found that the condition ensures an event occuring at the set of its initial contemporaries as its first instant. For a mathematico-logical treatment, a footnote refers to Wiener s [49]! (iii) In [49, p. 447f.], Wiener proves that, if an event satisfies the condition, then the set of its initial contemporaries is a maximal antichain containing that event (i.e., what Russell claimed to have shown). In advance, Wiener ascribes to Russell just having formulated the condition. What both Russell and Wiener tell concerning priority could perhaps be reconciled by another possibility, viz., that Russell did prove the equivalence of his condition on an event to the latter having a first instant earlier than Wiener, but without publishing his formal work at that time; while Wiener s [49] proof followed quite different lines. This guess rests on the footnote of [49, p. 441], where Wiener tells that the paper (as a whole) resulted from attempting to simplify and generalize what Russell had done so far. The fact that Russell in [43] referred wrongly to his [42] then might be some inaccuracy or slip of memory, which I, however, hardly can conceive in the light of the editor s (R. C. Marsh s) Preface to the collection to which [43] belongs: Lord Russell has been consulted on all

13 7 Solutions. Non-constructive 34 assumptions used. As a whole, the ensuing answers to the questions asked in sections 4 and 5 assume the axiom of choice or, at least, the principle of countable choice 35 and that every antichain (in our E, P ) of at most 2 elements extends to a maximal antichain. Theorem 1, however, needs the latter assumption for singletons in place of antichains only, and non-voidness of L (Section 3) merely means existence of a maximal antichain with no regard to what elements of E it should contain. I call E, P n-maximizing if every antichain of at most n elements extends to a maximal antichain; the assumptions mentioned before refer to this for n = 2, 1, 0, respectively. 36 Moreover, the condition of Theorem 3 characterizing separability of L, T implies that E, P is 2-maximizing and thus (whenever assumed) removes any reliance on a non-constructive principle as far as existence of maximal antichains is concerned; in particular, no maximizingness assumption is involved in sufficiency of the characterization of real intervals. Now, I believe that E, P is 2-maximizing or any of some other consequences of the axiom of choice discussed in Subsection 8.2 below does not imply the principle of countable choice. In case this belief is correct, it is important to note that the latter principle is used for both directions of Theorem 3. Under the previous separability assumption on E, P it remains the only non-constructive principle underlying one direction of Theorem 3 and Corollary 2 I name these assumptions and the directions in which they are used in parentheses near each statement below. Question 5 is most easily answered, so start with Theorem 1. L has more than one element if and only if P. (This uses just that E, P is 1-maximizing for if.) Answers to Questions 4a and 4b are slightly more difficult: Theorem 2. (a) L, T has a least element if and only if E has an SP -minimal element contained in ( having a first instant, i.e.). (b) (dually to the previous) L, T has a greatest element if and only if E has a P S-maximal element contained in + ( having a last instant, i.e.). matters relating to the text of the papers, and these, to the best of my knowledge, are here issued in the form which he wishes to be taken as final and definitive. It might be noteworthy in this context that [43], before having appeared in 1956 in the collection I am referring to, originally appeared in 1936 in the Cambridge Proceedings; while its references [42] (1st ed.) and [49] had appeared in 1914. For distinguishing Russell s from Wiener s credits in general, cf. footnote 7 above. 34 What constructive and non-constructive mean here and below is rendered more precise in Subsection 8.1. For a first approximation, constructivity avoids the axiom of choice and any weak variant of it. 35 Cf. [29, pp. 167, 2.1]. However, this principle even has been accepted in (parts of) constructive mathematics [45]. 36 Choose an SP -maximal and a P S-minimal element from an antichain to see that E, P is n-maximizing for every non-negative integer n as soon as it is 2-maximizing.

14 7 SOLUTIONS. (Both (a) and (b) use that E, P is 2-maximizing for only if.) In the following, situations like a P c S d P b are mentioned repeatedly. Such a situation is illustrated in Figure 2. (a P b follows by (I).) In such a b c d direction of time Figure 2: Antichain {c, d} in gap of chain {a, b}. a situation, {c, d} is an antichain in the gap of the chain {a, b} (where a chain is a subset of E forming a loset when P is restricted to it). If E were just {a, b, c, d}, L would be {{a, d}, {c, d}, {b, c}}, and T would do {a, d} T {c, d} T {b, c}, so {c, d} would be an instant T -between all the instants containing a in one direction and all the instants containing b in the other. This enables answering Question 3 as follows. Theorem 3. L, T is separable if and only if there is a countable subset D of E such that for all a, b P at the same time (i) if a + and b, then there are c, d D such that a P c S d P b; (ii) if a / + then, whenever e P S a, there are c, d D such that e P c S d P b; (iii) (dually to the previous condition) if b / then, whenever b SP e, there are c, d D such that a P d S c P e. (This uses countable choice in both directions and E, P is 2-maximizing for only if. In proving the theorem, I will generalize it to infinite cardinalities of dense subsets, then using the full axiom of choice.) Here is one by-product announced in the beginning: Corollary 1. L, T is dense (i.e., L is dense in L, T ) if and only if for all a, b P such that a + and b there are c, d E such that a P c S d P b (compare first condition in the above theorem). (This uses that E, P is 2-maximizing for only if.) The criterion in terms of E, P for L, T being dense given here is a weakened version of a condition stated by Russell in a footnote of [42, Lecture IV] and by Wiener in [49, pp. 446f.], 37 viz., for all a, b P, there are c, d E such that a P c S d P b. (III) 37 Both called according to the terminology of Principia Mathematica [48, 270] compact what nowadays is called dense (or dense-in-itself ); only Anderson in [1, pp. 256f.] still sticks to that historical term which was attaining a very different meaning in mathematics (topology) at the same time going back to a 1906 thesis by M. Fréchet, cf. Hausdorff 1914 [20] and [13, Chapter V]. According to this meaning, compactness of time would imply that there is a very first and a very last instant.

15 This, of course, is only a sufficient, no necessary condition for L, T being dense. 38 Wiener in [49, pp. 447ff.], however, uses an additional premise to prove that L, T is dense, viz., = E (in my terms, meaning every event has a first instant as indicated in Section 6 above). Thus he does not consider (III) sufficient as I do. But Wiener uses = E only to prove that the postulated two-element antichain in the gap of a two-element chain extends to a maximal antichain. By contrast, the present paper assumes maximal-antichain extendibility of any two-element antichain ( E, P 2-maximizing) from the start, so the crucial step in deriving denseness of L, T from (III) as mentioned goes without invoking Wiener s (in fact, Russell s) 39 additional premiss = E. In [43], Russell presents a condition on E, P necessary for L, T being dense and a pair of conditions he claims to be sufficient. A simple counter-example presented by Anderson [1, pp. 256f.] shows that this claim is wrong. 40 As a remedy, Anderson suggests 38 To see that (III) is no necessary condition for L, T being dense, let for every pair z 1, z 2 of integers I(z 1, z 2) be the real interval { r R z 12 z 2 2 z2 1 < r z 12 z 2 }. (Note that { I(z 1, z 2) z 1 integer } is only half of R for each z 2.) Then let E be the set of all these intervals and let P be the relation on E naturally deriving from <. Now, e.g., I(1, 0), I(1, 1) are P -neighbours; but L, T turns out to be isomorphic to R, < (map each real number r to the set of events containing r) and so is dense, since R, < is dense. (No event as chosen like this has a first instant, since there are no initial contemporaries ending earlier, but lots of contemporaries beginning later. Replacing in the definition of I(z 1, z 2) by < would make no difference but the need to recognize that, now, each I(z 1, z 2) is element of that maximal antichain which contains those events that include z 12 z 2.) Continuing the example (forget about the variation just mentioned) shows that E, P needs not to be K-dense (cf. [6, p. 76]), which following [19] also has been called CAC (chain-antichain-complete), in order that L, T be as wanted: { I(2 n 1, n) n positive integer } is a chain. By the axiom of choice but also in a constructive way it can be extended to a maximal chain. But no such extension intersects with the maximal antichain of events including 1. (One constructive way of extending the chain to a maximal antichain would be adding all I(2 n 1, n + m) for further positive integers m; if C then is the chain obtained so far, { { r + z r M } M C, z integer } is a maximal chain. It intersects with no maximal antichain that corresponds to an integer.) 39 Wiener [49, p. 447] attributes the condition to Russell. 40 One of the conditions Russell [43] claims to be sufficient is just that E, P is 2-maximizing the surrogate for the axiom of choice which the present paper uses as well. The other condition is that no event lasts only for an instant. This was Russell s epistemological starting point already in [42, Lecture IV] (cf. his later [44, p. 293]). It implies the first-order condition SP S where = { a, b E E a = b } and SP S is the composition of relations according to the well-known notation (cf. [17, p. 3]) like SP and P S and below in Subsection 8.3 P SP. The converse implication holds when E, P is 2-maximizing (Russell s first condition). SP S has just nothing to do with linear denseness of L, T i.e., these two conditions are independent (for the direction not dealt with by Anderson, take the rational numbers for E and T = < (E E); then L, T = E, P is dense, though not SP S here S =.) It should not be supplemented as in Kleinknecht [26], but just ignored. (Even less one should add an assumption like [26, A1.77] which contradicts Russell s no event lasts only for an instant Kleinknecht assumes that the members of each antichain t have a common part c; but if t is a maximal antichain, this event c would last for its only instant t, since all d, d overlap-

16 7 SOLUTIONS. for all a, b P, there is c E such that a P c P b as an alternative to the conjunction of (III) and = E for entailing that L, T is dense obviously not realizing that under his [1, p. 255] assumption of the axiom of choice (which entails E, P 2-maximizing, see Subsection 8.2 below) (III) alone is a condition entailing that L, T is dense and a (much) weaker one than his own proposal. Thus, to my knowledge, the statement of the previous corollary is new. 41 To tackle Question 2, define for every A E U(A) := { e E a P e for all a A }; H(A) := { e E e P b for all b U(A) }. U(A) comprises all those events that are wholly preceded by all events of A. H(A), by contrast, is some kind of hull of A collecting not only all those events that do not end later than some event from A (so, of course, A is a subset of H(A)), but collects even those events that do end later than all events from A, but only non-uniformly. 42 H(A), U(A) is a pair of maximal sets of events such that all the events of the first set wholly precede all the events of the second set. Every pair with this property is ping with c overlap with all members of t [26, D1.43], so d, d t and d S d. Instead of contradiction one might derive non-existence of events; this, however, would be no better philosophically, while, more formally, Kleinknecht seems to use a classical second-order logic where first-order variables range over events; classically, as opposed to free logic, existence of some event then obtains as a logical truth.) Russell s [43] flaw looks like this: he correctly states that SP S (consequence of no event lasts only for an instant) implies his necessary condition P SP SP S (notation for compositions of relations like previously; Russell s wording: reduces the latter condition to a true one; he explicitly acknowledges that this necessary condition is not sufficient). He provides no indication how this, together with his first condition (axiom-of-choice surrogate), could help intermediate instants to exist. Perhaps he believed that, for 2-maximizing E, P, a s L, b t L, a P b, and a SP c S d P S b ( witnesses c, d for a SP SP S b according to P SP SP S), {c, d} extends to a maximal antichain u such that s T u T t. Anderson s example, of course, also shows that this belief is wrong. Or could the following spurious idea have underlied the thought: if Ψ (P SP SP S) is necessary for X (denseness of L, T ) and Φ (no event lasting for an instant only) implies Ψ, then Φ is sufficient for X as well (if Φ Ψ and X Ψ, then Φ X)? Anyway, it is not quite appropriate to say as Anderson [1, p. 256n.] does there were an error in Russell s proof, or even to point at a single line. The line to which Anderson points is just the first one where Russell s claim occurs, after some zigzag which Russell seems to consider its proof. More formally/precisely than in Anderson [1], E, P is a counter-example to Russell s [43] claim when E is the set of integers and T = { a, b E E a + 1 < b } (so S = { a, b E E a b { 1, 0, 1} }). 41 [18, p. 170] tell that Theorem 3.4 of [17] provides access to conditions that are necessary and sufficient, but the matter is rather technical and we shall not pursue it here. I estimate Corollary 1 and its proof as much more simple than that theorem, therefore it seems unlikely that [18] could have meant the condition of my corollary. 42 I.e., b H(A) if a P c S b for all a A and some c depending on a, without there being a single c such that a P c S b for all a A. It is also easy to see that the operator H is idempotent and monotone.

17 H(A), U(A) for some A E, and the property generalizes the notion of a Dedekind cut (in the sense of [40, Definition 2.22], e.g.). 43 My answer to Question 2 then is Theorem 4. L, T is complete if and only if every A E satisfies one of the following conditions: 44 (i) A or U(A) is empty; (ii) H(A) has a P S-maximal element contained in + ; (iii) U(A) has an SP -minimal element contained in ; (iv) for every a H(A) there is some c neither in U(A) nor in H(A) such that a P c, and for every b U(A) there is some d neither in U(A) nor in H(A) such that d P b. (That E, P is 2-maximizing is used for only if again.) To get an idea of what goes on in cases (ii) and (iii), the reader may look at Theorem 2. If one of them applies, the boundary of one of U(A) and H(A) generates a bound as desired for both of them, unless the events outside of U(A) and H(A) according to case (iv) form a maximal antichain that is a bound as desired. 45 The previous theorems solve the problems formulated in Section 4 as explained in Section 5. This may still not be perfectly clear, so I give two examples in the ensuing corollary. Its first part deals with one class of instances of Question-Scheme 1* the one Thomason seems to be interested exclusively, namely (up to isomorphism) just Question 1. The second part deals with Question 1 + which appears interesting to me. 43 That is why they came to my mind when I looked for a characterization of L, T being complete. By [40, Lemma 2.23], for losets completeness is equivalent to Dedekind completeness, where the latter means that one of the members of every Dedekind cut contains a boundary of itself. Actually, these pairs together with the complement of the union of their members form just the triples that are instants of time in the sense of Walker s construction according to [46]. 44 One might wonder whether an answer to a question in terms of E, P may quantify over sets of events. However, the problem needs too much words of explaining, so I only give the following hints to my thoughts about it: (i) The elements of L are, admittedly, sets of events but special ones. (ii) To characterize a second-order property of L, T in terms of E, P, surely a second-order property of E, P should be allowed (complexities should be allowed to match). 45 Why did I put case (iv) so much more clumsy than this explanation? Answer: in order to stay in terms of E, P. Otherwise, the theorem would not truely be an answer to Question 2, in terms of E, P. For the same (or respective) reason, I did not define e by saying the set of maximal antichains that contain e has a least element (as first instant, i.e.). One could object that after having quantified over subsets of E there is no more point in such asceticism. I know, however, at least one possibility to counter this objection in some terms of complexity. The formulation was inspired by the definition of a complex open gap in [17, p. 42].

18 8 PROOFS. Corollary 2. (a) L, T is isomorphic to I, < for some non-void open real interval I (I = R, e.g.) if and only if E neither has an SP -minimal element contained in nor a P S-maximal element contained in +, there is a countable subset D of E according to Theorem 3, and every A E satisfies one of the cases of Theorem 4. (b) L, T is isomorphic to I, < for some non-trivial real interval I if and only if there is a countable subset D of E according to Theorem 3, every A E satisfies one of the cases of Theorem 4, and (if E has an SP -minimal element in and a P S-maximal element in + ) 46 P. (Both (a) and (b) use countable choice and, for only if, that E, P is 2-maximizing.) 8 Proofs. Whereas so far I have tried to explain everything mentioned, the present section presumes some basic mathematical capabilities and, in places, some basic knowledge of axiomatic set theory as can be gained from, e.g., [27, Chapter 1] or [29]. 8.1 Constructivity as opposed to choice principles. This subsection starts elaborating the remarks beginning Section 7 on what the paper particularly assumes. I explain what I claimed there on avoidability of choice and similar principles as well as the loose use of terms like constructive. Neglecting a small amount of terminology used in parts of subsections 8.2 and 8.6, a reader used to apply the axiom of choice with the same ease as any other common set-theoretical principle may skip the subsection without loss. The axiom of choice AC for short seems, in general, not to be questioned in nowadays mathematics and is considered part of standard (axiomatic) set theory. 47 On the other hand, when it found broad mathematical attention for the first time nearly a hundred years ago, 48 adversaries included Peano, Borel, Lebesgue, and Baire see [28, pp. 103ff.] or, for more, [34]. Russell only temporarily was convinced of it ([1, p. 255]) and considered special assumptions on events to guarantee existence of instants without its use (see Subsection 8.2). In the meantime, there have, at least, always been mathematicians who liked to point it out when a proof used AC or similar non-constructive existence claims. 49 Moreover, there is still some interest in 46 Otherwise P holds anyway. 47 Accordingly, [1, pp. 254ff.] and [46, p. 87] use it for a modern treatment of instants along Russell s lines. 48 However, there was a prehistory of a quarter of a century cf. [34, pp. 5ff.]. 49 At least, it seems to be common practice at Munich University to tell first-term students of mathematics that existence of bases for all vector spaces is due to Zorn s

8.1 Constructivity as opposed to choice principles. 19 finding proofs only using constructive existence claims, even with an intuitionistic intent here I am alluding to constructive mathematics, see, e.g., [8] or [45]. Subsections 8.1 and 8.2 address such interests. In axiomatic set theory, everybody knows what set theory without the axiom of choice is. I designate this part of set theory by ZF. 50 Subsections 8.1 and 8.2 tell something about which claims of the paper hold by means of ZF only. The latter phrase will return to stress this. Whenever the paper claims that something holds constructively or that a proof is constructive, this just means that it is provable by means of ZF only. This terminology is, definitely, not correct considering what may be understood by constructivity in mathematics. 51 I consider it justified, however, by the fact that purportedly constructive proofs avoid use of (ZF-)equivalents of AC ([41, p. xiv]) or of weak versions of it (other choice, maximality, or extension principles, some of which will be reviewed below) by presenting definitions the only free variables of which stand in place of things which exist by the hypothesis of the implication to be proved. 52 This is to indicate how a formal 53 proof could work. 54 Thus, when the hypothesis of a theorem asserts existence of some x such that... and claims respective existence of some y such that..., a definition of such a y in terms of x 55 suffices for constructivity for holding by means of ZF only, i.e. There is no need to specify a respective x (by any definition ). 56 According to the beginning of Subsection 7, certain claims of the paper hold by means of ZF only. Like everywhere in set theory, the corresponding metamathematical claims will not be checked in a strict manner. The informal proofs only implicitly indicate their correctness. The specifically Lemma. And teaching the Lebesgue-measure involves telling that it is due to AC that the Lebesgue-measure cannot be defined for all bounded sets. 50 I prefer to consider ZF a purely formal theory, i.e., a certain set of formulas of some formal language of first-order predicate logic with identity, defined by a finite list of axioms and axiom-schemata; cf. [14, p. 13] or [27, p. xvi]. In the present paper, however, I cannot distinguish this formal theory from the body of informal theorems and proofs which could be formalized in proper ZF according to common practice (cf. [27, pp. 1f.]). 51 Constructive mathematics typically rejects the principle of tertium non datur which is not questioned in classical mathematics cf. [45]. I will not check whether any proof invokes this principle. The paper cannot address the question what constructive really means. To my knowledge, no precise answer is agreed upon by a majority of authors, cf. [8] and [45]. 52 In spite of this justification, I will from now on prefer by means of ZF only to constructive. 53 Cf. footnote 50 above. 54 My proofs of this kind may be acceptable for one or the other constructive mathematician I hope so, but I have not checked this. 55 In the corresponding formal way, there typically is a defining class term containing no free variable but x; cf. [29, p. 171]. 56 Peter Schuster (the author of [45] and in this respect an expert) thinks that this feature of relative constructivity of some proofs may be accepted by many, though not by all exponents of constructive mathematics.

20 8 PROOFS. metamathematical aspects of the claims of the paper are adressed in an intermediately explicit manner in Subsection 8.2, in parts of Subsection 8.6, and in the rest of the present subsection as follows. Corollary 2 rests on (some of) Facts 1, 1* and 1 +, and so do similar claims I alluded to. My corresponding metamathematical claims rely on Meta-Fact. Facts 1, 1* and 1 +, hold by means of ZF only. For those facts (on the object level ), I am referring to proofs to be found in the literature and to obvious variants of such proofs. For my corresponding metamathematical claim, one has to check that respective proofs can be carried out in ZF: Proof of Meta-Fact. In each case, an isomorphism as wanted arises as a straightforward (indeed: unique) extension of an isomorphism of countable dense subchains which have no ends, obviously involving nothing more than ZF. That the construction of such a core isomorphism as well can be carried out by means of ZF only is to my knowledge most explicit in [36, p. 32f.] and in [5, p. 200]. 57 8.2 Existence of Russellian instants of time. This subsection reviews some metamathematical facts relating the basic assumption of the paper (according to the beginning of Section 7) that E, P is 2-maximizing to AC and some of its consequences 58. Russell, doubting AC, proposed to avoid it by E = in [42] (like [49], see Section 7 above) and by E, P is 2-maximizing in [43] (see footnote 40 above) in order to guarantee existence of enough instants. 59 These pains and 57 Other expositions might appear at first glance to use the principle of dependent choices ([2, pp. 176f.], e.g.; cf. [29, pp. 168f.], [45]). Like in Subsection 8.2, however, countability allows to well-order the codomain and so to define a map cf. Russell s illustration by pairs of shoes vs. socks (see [22, p. 351]). This is obvious for all the proofs I have seen, but the recommended proofs in [5] and in [36] are the only ones which make the definition explicit. The latter, however, might be supplemented by the hint that f 1 there is a second map which is constructed simultaneously with f and only in the end turns out to be the inverse map of the latter. 58 Consequences by means of ZF only, i.e., as explained in Subsection 8.1 above. 59 In [43], Russell discusses E = again, as well E = + which serves the same purpose, viz., giving an instant to each event to be at ; moreover S \ SP S, which guarantees that there is any one instant at all (it is the negation of Thomason s [46] condition corresponding to denseness of the Walker construction). Since he neither believes in AC nor in any of his own conditions, he remains sceptical about existence of instants. He might have been happy to know that instants may be constructed in other ways than as maximal antichains. Such alternatives are generalized Dedekind-cuts as in [46] or the entirely different, though surprisingly straightforward way of [17, Sections 7.5f.]. I propose this version of generalized Dedekind-cuts : For a E, let h(a) := { e E not a P S e }. Then K := { t E h(a) t whenever a t } might be considered the set of instants, and an event a might be considered at ([42, Lecture IV]) an instant t if t h(t), but